Synthetic Audio Generation Framework for Air Traffic Control Speech Recognition

A new synthetic data generation framework for improving Automatic Speech Recognition (ASR) in Air Traffic Control (ATC) has been proposed, addressing challenges like channel noise and non-native accents. The approach utilizes a combination of neural generation techniques, including Text-to-Speech, Voice Conversion, and a novel L1-to-L2 accent conversion framework. Experiments with the Whisper model on the ATCO2 corpus showed significant reductions in word error rates when fine-tuned with synthetic data, highlighting its potential for enhancing ASR performance in noisy, specialized domains.